Claude review: drm/gem/shmem: Introduce __drm_gem_shmem_free_sgt_locked()

[Claude Code Review Bot <claude-review@example.com>]

Patch Review

Extracts the sgt teardown logic into a new exported function. The refactoring is clean and the function is correctly guarded with `dma_resv_assert_held()`.

**Naming mismatch in kdoc**: The kdoc comment header says `__drm_gem_shmem_release_sgt_locked` but the actual function is named `__drm_gem_shmem_free_sgt_locked`. The changelog mentions fixing this in V10, but the mismatch is still present:

```c
/**
 * __drm_gem_shmem_release_sgt_locked - Unpin and DMA unmap pages, and release the
 * cached scatter/gather table for an shmem GEM object.
```

vs:

```c
void __drm_gem_shmem_free_sgt_locked(struct drm_gem_shmem_object *shmem)
```

This should be `__drm_gem_shmem_free_sgt_locked` in the kdoc to match.

**No NULL check on shmem->sgt**: The original code in `drm_gem_shmem_release` checks `if (shmem->sgt)` before calling the teardown. The new `__drm_gem_shmem_free_sgt_locked()` does not check for NULL and will crash if called with `shmem->sgt == NULL`. The call site in `drm_gem_shmem_release` preserves the NULL guard, but external callers (the Rust bindings in patch 4) need to be careful. The Rust side does call this only after a successful `drm_gem_shmem_get_pages_sgt_locked`, so it should be fine in practice, but a defensive NULL check (or at least a `WARN_ON(!shmem->sgt)`) in the exported function would be safer.

**Kdoc says "Unpin" but function doesn't unpin**: The kdoc says "Unpin and DMA unmap pages" but the function only does DMA unmap + free the sgt structure. It does not call `drm_gem_shmem_put_pages_locked()`. This is accurate to the code behavior (the caller in `drm_gem_shmem_release` calls `put_pages_locked` separately), but the kdoc is misleading.

---
Generated by Claude Code Patch Reviewer